Document Diffusion artifacts

README.md

@@ -5,18 +5,24 @@ tags:
 - ddpm
 - cifar10
 ---
-# Diffusion training outputs
+# Diffusion final checkpoints
 
 Unconditional and conditional DDPM training artifacts for CIFAR-10, using
 custom UNet architectures with attention and EMA.
 
-Each experiment directory contains PyTorch checkpoints, generated sample images,
-and a training history CSV. Checkpoints contain model, EMA model, optimizer,
-scheduler, epoch, and loss state. Use the matching code from the Diffusion
-project to load them; these are custom PyTorch models.
+Only the final unconditional and conditional checkpoints are published:
+
+- `ddpm_with_attention_ema/checkpoints/ddpm_final.pt`
+- `conditional_ddpm_with_attention_ema/checkpoints/conditional_ddpm_final.pt`
+
+Each file is about 3.11 GiB, about 6.2 GiB combined. Intermediate checkpoints,
+sample images, and training history CSV files are not published. The original
+full final checkpoints contain model, EMA model, optimizer, scheduler, epoch,
+and loss state. Use `ema_model_state_dict` for EMA inference weights, with the
+matching custom architectures from the code repository.
 
 Download the repository contents into `Unet/outputs` in the code project.
 
 ## Code
 
-Training code and setup instructions: https://github.com/wangwenbinw/Diffusion
+https://github.com/wangwenbinw/Diffusion